Earth Simulator Center

ESC seminar No. 56

Adaptive localization and adaptive inflation methods with the LETKF

Date

Jun. 9 (Wed.), 2010, 14:00-14:45

Place

Meeting room 1&2, 2F Conference Building, YES, JAMSTEC

Speaker

Takemasa Miyoshi (Research Assistant Professor, Department of Atmospheric and Oceanic Science, University of Maryland)

Language

English or Japanese

Abstract

In ensemble Kalman filters with high-dimensional systems, the error covariance estimated by ensemble members is usually rank deficient and contains inevitable sampling errors. Moreover, it is known that the error covariance is underestimated due to various sources of imperfections such as the limited ensemble size and model's nonlinearity. These are the typical reasons of filter divergence and are usually treated by empirical covariance localization and inflation techniques. However, it is prohibitive to optimize the parameters of localization and inflation, since they would depend on geographical locations, vertical levels, and even seasons. In this study, adaptive methods for covariance inflation and localization, which are particularly efficient with the Local Ensemble Transform Kalman Filter (LETKF), are proposed. The methods are tested with a low-resolution AGCM and show promising results.

Contact

Takeshi Enomoto
Earth Simulator Center
TEL: 045-778-5867
e-mail: