A coupled data assimilation (CDA) system based on the dimension-reduced projection four-dimensional variational (DRP-4DVar) method has been developed for FGOALS-g3. The system assimilates monthly ocean temperature and salinity analyses using one-month assimilation windows, producing ocean states that are more consistent with observations and better capture major climate variability modes, including the El Niño–Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), and Atlantic Multidecadal Oscillation (AMO). Using initial conditions generated by the CDA system, a suite of 10-year hindcasts is initialized annually from 1961 to 2013. High decadal prediction skills are achieved for global mean surface air temperature anomalies, PDO, AMO and East Asian summer monsoon. The skills are competitive with or superior to those of most CMIP6 decadal prediction systems. These results demonstrate the effectiveness of DRP-4DVar-based CDA in initializing FGOALS-g3 and improving decadal climate prediction.