In order to explore the changes of terrestrial water storage (TWS) and its components in China from April 2002 to March 2021, the interval data between GRACE gravity satellite and its follow-on satellite GRACE-FO were interpolated by using a hybrid VMD-LSTM model based on the decomposition-integration idea. The Theil-Sen slope analysis method and the Mann-Kendall trend test method were used to study the spatiotemporal evolution of TWS and its components in nine major river basins in China. The random forest method was used for analyzing the relative contribution rate of each component to TWS. The results showed that the VMD-LSTM model can effectively interpolate the GRACE sequence in China, with the Nash-Sutcliffe efficiency coefficient greater than 0.6, correlation coefficient greater than 0.9, and root mean square error less than 2cm in most regions, which significantly improves the interpolation accuracy. There is a spatial consistency between the change trends of TWS and groundwater storage. Except for the Southeast River Basin, Pearl River Basin, and Yangtze River Basin, TWS in other basins shows a downward trend, which is mainly caused by groundwater deficit. TWS in western arid and semi-arid regions and North China is mainly affected by groundwater storage, and TWS is greatly affected by canopy water storage and soil water storage in humid and semi-humid regions, compared to arid and semi-arid regions.