[p1] Non-Gaussianity and nonlinearity


[p1-14]

Kernel regression approach for ensemble Kalman filters

S. Nakano (The Institute of Statistical Mathematics)

 
Abstract

The ensemble Kalman filters are now widely used for data assimilation in nonlinear systems in various field. In the ensemble Kalman filters, the uncertainty in a system is represented by a set of possible scenarios called ensemble. An estimate of the system state is produced by a linear combination of the ensemble members. This procedure of the ensemble Kalman filters can be rewritten in a form of the kernel regression approach. A formulation based on the kernel regression approach enables us to allow a nonlinear relationship between the state and the observation. In this study, a formulation of the ensemble Kalman filters based on the kernel regression approach is introduced and some extentions of the ensemble Kalman filters are discussed.