Multiyear climate prediction with initialization based on 4D-Var data assimilation

An initialization relevant to interannual-to-decadal climate prediction has usually used a simple restoring approach for oceanic variables. Here we demonstrate the potential use of four-dimensional variational (4D-Var) data assimilation on the leading edge of initialization approach particularly in multiyear (5 year long) climate prediction. We perform full-field initialization rather than anomaly initialization and assimilate the atmosphere states together with the ocean states to an atmosphere-ocean coupled climate model. In particular, it is noteworthy that ensembles of multiyear hindcasts using our assimilation results as initial conditions exhibit an improved skill in hindcasting the multiyear changes of the upper ocean heat content (OHC) over the central North Pacific. The 4D-Var approach enables us to directly assimilate a time trajectory of slow changes of the Aleutian Low that are compatible with the sea surface height and the OHC. Consequently, we can estimate a coupled climate state suitable for hindcasting dynamical changes over the extratropical North Pacific as observed.