DescriptionThe initialization for decadal predictions using a coupled model includes a long-term cycle of data assimilation in the coupling framework, which is an essential step of decadal prediction. The skill of decadal prediction closely relies on the performance of the assimilation scheme used in the initialization. Currently, most decadal prediction experiments conducted in the Coupled Model Intercomparison Project Phase 5 (CMIP5) simply used the nudging approach, and very few applied the variational or ensemble methods that are very expensive in computing. Therefore, it is necessary to develop an assimilation scheme that can well use the observations in an economical way. In this paper, we propose a coupled assimilation scheme that can incorporate 4-dimensional ocean observations (say temperature and salinity) into the IC of ocean component model and in the meantime can be timesaving. Numerical experiments using a coupled climate system model are conducted to test the performance of the new assimilation scheme.