JpGU-AGU Joint Meeting 2017

Presentation information

[EE] Oral

A (Atmospheric and Hydrospheric Sciences) » A-AS Atmospheric Sciences, Meteorology & Atmospheric Environment

[A-AS12] [EE] High performance computing for next generation weather, climate, and environmental sciences using K

Sat. May 20, 2017 9:00 AM - 10:30 AM 101 (International Conference Hall 1F)

convener:Hiromu Seko(Meteorological Research Institute), Takemasa Miyoshi(RIKEN Advanced Institute for Computational Science), Chihiro Kodama(Japan Agency for Marine-Earth Science and Technology), Masayuki Takigawa(Japan Agency for Marine-Earth Science and Technology), Chairperson:Hiromu Seko(Meteorological Research Institute)

10:00 AM - 10:15 AM

[AAS12-05] Perturbation Methods for Ensemble Data Assimilation

*Kazuo Saito1,2, Masaru Kunii1, Duc Le2,1, Takuya Kurihana3 (1.Meteorological Research Institute, 2.Japan Agency for Marine-Earth Science and Technology, 3.University of Tsukuba)

Keywords:ensemble data assimilation, ensemble transform, perturbation method

Ensemble data assimilation methods are widely noticed as the analysis methods suitable for HPC which have potential to improve the accuracy of numerical weather prediction, substituting for or combining with the variational methods. Ensemble data assimilation methods have an advantage in terms of the development cost over the 4-dimensional variational method in that the adjoint models are not necessary, however, their performances are still arguable and likely have room for further improvement.
In ensemble data assimilation, the forecast error, which is necessary in data assimilation, is estimated by perturbations of the ensemble forecast, while characteristics of the ensemble forecast strongly depend on how the initial ensemble was generated. The ensemble transform (ET), eigenvalue decomposition of the analysis error covariance matrix, is widely used as the initial ensemble perturbation generator for the most ensemble data assimilation including ensemble Kalman filter such as LETKF and the ensemble variational method (EnVAR). The ensemble transform has an advantage in that the magnitude of perturbations (initial ensemble spread) can reflect the magnitude of the analysis error, but on the other hand, it is known that the growth of the errors is slower than other methods such as the singular vector method and the BGM method. In the previous studies for the mesoscale ensemble system (e.g., Saito et al.; 2011; 2012), perturbations from LETKF were not necessarily better than other methods as the initial perturbations, which may affect the accuracy of the analysis field. Non-diagonal components in the transform matrix likely contaminate the synoptic scale structure of the bred vectors in the ensemble forecast in the assimilation window when the localization is applied.
We started to tackle this problem, and in the presentation, some preliminary results using SPEEDY-LETKF will be shown, including spatial strucure and power spectrum of ensemble perturbations by diaonal and off-diagonal components of the transform matrix.