AbstractThe East Asian summer monsoon (EASM) has undergone significant decadal variations since the late 1970s, which greatly affects regional economic and societal activities. However, most available EASM predictions are limited to seasonal to interannual scales, with indistinctive improvements. This study significantly improves decadal prediction skills of the 10‐year‐averaged EASM index (EASMI) and related decadal variations of the East China summer precipitation anomalies (ECSPAs) in the 10‐member decadal prediction experiments started once a year from 1960 to 2005 using a coupled climate model. These improvements are mainly attributed to the use of more realistic and well‐balanced initial conditions (ICs) from a weakly coupled data assimilation system that incorporates the monthly mean atmospheric reanalysis data into the model using the dimension‐reduced projection four‐dimensional variational data assimilation approach, a kind of four‐dimensional ensemble‐variational hybrid method. The atmospheric initialization ameliorates the ICs of Pacific Decadal Oscillation (PDO) through the air‐sea coupling, and thus leads to better PDO hindcasts and forecasts (under the representative concentration pathway 4.5). The improved PDO hindcasts (forecasts) result in more accurate hindcasts (forecasts) of land‐sea thermal contrast, and thus have significant impacts on the EASMI. The improved decadal predictions of the EASMI then contribute to the ameliorated prediction on decadal variations of the spatial distribution of ECSPAs. This study highlights the importance of ICs for a coupled model in reasonably presenting decadal variabilities of the EASMI.