Abstract: The North Atlantic Oscillation (NAO) is one of the major modes of atmospheric circulation over the North Atlantic in winter, and its interannual variabilities play an important role in climate variabilities over many regions of the world. However, the skills for its prediction are not good enough at present. In this paper, a weakly coupled data assimilation system based on global observational atmospheric data is established using the Dimensional-Reduced Projection Four-Dimensional Variation (DRP-4DVar) assimilation method, which can directly assimilate monthly mean reanalysis data. The results of decadal hindcast experiments indicate that this system can significantly improve the hindcast effects of interannual variabilities in wintertime NAO and interannual variabilities of related winter surface temperature over northern Europe, the eastern United States and northern Eurasia, and the correlation coefficients are all above the 0.05 significance level t-test at least. These improvements are mainly attributed to the freely developed air-sea coupling in the coupled model that can store atmospheric observations in its ocean component, and thus improve the spatial and temporal variabilities in the "tripole" pattern of the Sea Surface Temperature (SST) distribution over the North Atlantic and related "tripole" SST temporal variabilities. This study emphasizes the importance of the accuracy of initial states of the coupled model in improving the hindcast abilities of the coupled model on the simulation of wintertime NAO interannual variabilities.