The terrestrial water budget over northern China, which plays an important role in water resource management, has experienced great changes during the past decades. However, its spatiotemporal variations in the past calculated from individual datasets remain quite uncertain. In this study, we improve the accuracy of evapotranspiration (E), precipitation (P) and runoff (R) estimates by merging remote sensing, reanalysis, data assimilation datasets and ground observations, and further analyze the spatiotemporal characterization of the terrestrial water budget at 0.25° over northern China during the period of 1984–2010. The results illustrate that using any of the individual datasets, there is significant uncertainty and an obvious seasonal cycle in the terrestrial water budget. Large differences exist among the different datasets, and the merged E, P and R outperform the individual datasets. The root mean square errors (RMSEs) from cross-validation are 8.4–14.2 mm, 15.9–27.3 mm and 4.1–14.2 mm for the monthly merged E, P and R at the site scale of the different basins, respectively. The spatial patterns of the merged annual E and R are consistent with that of P due to the water limitations mainly controlled by P. The interannual variations in these hydrological variables indicate a slight increase in the variables from 1984 to 1998, with a large El Niño event, and a larger decline thereafter as a result of a large-scale drought. However, decadal trends in terrestrial water storage changes (TWSC) over all five basins inferred from the merged products tend to increase to some extent with climate warming over the studied time period. The Budyko curve reveals that an increase in vegetation coverage increases the evaporation ratio (E/P) to some extent, but climate change is the dominant driver for the variations in the hydrological variables in these regions.