[9-1] February 28, 14:50-15:20

[Invited]

A local hybrid data assimilation method for surface observations using vertical correlation information and flow-dependent error covariance

Shu-Chih Yang (National Central University, Taiwan) and Jeff Steward (University of California, US)

It is widely recognized that hybrid data assimilation systems combining the ensemble Kalman filter (EnKF) and the variational approach (ND-Var, where N = 1,...,4) perform better than either system alone. In this work we develop a simple local ensemble transform Kalman filter (LETKF) and local variational hybrid and use these to perform a case study of assimilating surface observations. The impact of surface observations is limited with the ensemble-based method alone due to the presence of sampling errors in the vertical. Vertical localization, often employed to counter sampling error, results in inconsistent vertical structure. In contrast, a local variational method which uses column principal components of a static (i.e. no errors-of-the-day) full-rank background error covariance creates consistent vertical updates. A hybrid method is required in order to fully take the advantage of the horizontal flow-dependent background error statistics from EnKF and the robust vertical correlation information from the static background error statistics. This concept is first proven with idealized assimilation experiments using a quasi-geostrophic model where the truth is known. Results show that when the analysis update can incorporate the vertical correlation information, the accuracy of the upper-level model states can be effectively improved and the filter divergence can be avoided. This local hybrid method is further demonstrated with an experiment that assimilates a single surface observation with the WRF model based on observations from Typhoon Dujuan in 2015. More results will be presented at the meeting.