New methods for an ensemble prediction system

Principal Investigator

Linus Magnusson
Department of Meteorology, Stockholm University
Department of Meteorology
Stockholm University
S-106 91 Stockholm, Sweden

linusm@misu.su.se

Other researchers: Erland Källén, Jonas Nycander

Project description

We are studying methods to improve initial perturbation techniques for ensemble prediction. The aim of our work is to compare how well different methods samples the uncertainties in the analysis and develop new or modified methods to obtain a better understanding of the error dynamics and improve the ensemble performance.

As a first part of the project, we compared the operational singular vector setup at ECMWF (Leutbecher and Palmer, 2008) with the breeding vector method used by e.g NCEP until 2006 (Toth and Kalnay, 1997). The breeding method was implemented at ECMWF yielding the possibility to compare both methods using the same forecast system. The main conclusion is that results in terms of skill scores show similarities but with a slight advantage for the singular vectors system in the northern hemisphere. The study is accepted for publication in Monthly Weather Review (Magnusson et al., 2008).

We have now continued our study by developing a new technique to generate stochastic initial perturbations. The method uses the structure of the difference between two randomly chosen atmospheric states as initial perturbations (RF method). The perturbations generated with this method do not use any information about the present atmospheric state, i.e. the perturbations are not flow dependent. The method has been implemented at ECMWF and in the ongoing study we compare the RF method with the operational singular vector system and the ET method, which is currently used at NCEP for generating initial perturbations (Wei et al. 2008). The preliminary results are promising for the new technique. The skill scores are comparable with both of the operational systems and yields in some cases better skill scores. However there are some undesirable properties in the RF perturbation method, and we believe that the results can be improved.

For the first part of our work we have used the member state computer allocation (Sweden).

As a next step in our project we are planning to improve the RF method. We have several ideas how this can be done. One method is to use spectral filtering to favour important scales. Another idea is to use a more selective process to choose the atmospheric states for the perturbation calculations.

We believe that this project could give new insights for the generation of initial perturbations for ensemble prediction. We are planning to collaborate with Martin Leutbecher at the research department at ECMWF. The results from our study could act as a benchmark of the operational EPS at ECMWF and the conclusions from our research could be used at the research department and hopefully help in improving the current ensemble system. Our new technique is also applicable to seasonal predictions and climate scenarios.

For all our experiments so far we have used the IFS-model with resolution T L255L40. This is chosen to be close to the operational resolution at ECMWF. We are running 10-day ensemble forecasts with 20 members. Each ensemble forecast needs about 2000 SBU. To obtain a statistically reliable dataset we expect to need 45 forecasts. To make it possible to run two different versions of the method including tuning of the systems we apply for 200 000 SBU. For the data storage for the experiments, we apply for 1020 GBytes.

References

Leutbecher, M. and T. N. Palmer, 2008: Ensemble Forecasting. J. Computational Physics, 227, 3515-3539.

Magnusson, L., Leutbecher, M. and E. Källén, 2008: Comparison between Singular Vectors and Breeding Vectors as Initial Perturbations for the ECMWF Ensemble Prediction System. Mon. Wea. Rev., In Press, doi: 10.1175/2008MWR2498.1

Toth. Z and E. Kalnay, 1997:Ensemble Forecasting at NCEP and the Breeding Method. Mon Wea. Rev., 125, 3297-3319.

Wei, M., Toth, Z., Wobus, R. and Y. Zhu, 2008: Initial perturbations based on the ensemble transform (ET) technique in the NCEP global operational forecast system. Tellus, 60A, 62-79.

Additional information

New Project for 2009

Would accept support for 1 year only.