Covariance inflation is used to prevent underdispersiveness in the ensemble-based data assimilation. It is prohibitive to tune the inflation coefficients manually in both space and time in a large dimensional system such as geophysical application. Several methods have been proposed to estimate inflation coefficients adaptively. Miyoshi (2011) proposed the adaptive multiplicative inflation scheme based on observation minus background statistics. Although this method has been successfully implemented in various systems, it turned out that this scheme does not work with the inaccurate observation error covariance. In this presentation, a new adaptive inflation scheme using the relaxation to prior spread method (Whitaker and Hamill 2012) will be proposed. This new scheme uses the relaxation to prior spread method for representing spatially varying inflation factor while its temporary varying coefficient is estimated from the observation based statistics. Preliminary result using a simple low resolution GCM suggests that this new scheme is robust in handling improperly tuned observation error covariance. Basic investigations of the new scheme and its performance compared to the existing inflation scheme will be shown.

---

*email: ota@met.kishou.go.jp
*Preference: Oral