Abstract: Objectives:  The time-varying gravity field model calculated from gravity recovery and Climate Experiment (GRACE) data provides a new method for the study of land water reserves. However, GRACE data can only calculate the overall water storage variation of a grid point, including surface water, soil water, ground water and vegetation water, but can not provide the deep groundwater composition at the vertical level.   Methods:  The research goal of this paper is to use the wavelet analysis to decompose GRACE signal which is deducted GLDAS hydrological model surface water composition, and use the decomposed wavelet sub function combined with the measured data of water wells in the basin to carry out regression analysis on the groundwater components, remove soil water and vegetation water, and obtain the regression coefficients of different wavelet subfunctions in the United States through two-dimensional surface interpolation, so as to reconstruct the long-term continuous change series of groundwater reserves.   Results:  The results show that the correlation coefficient of more than 61.84% of the test points is more than 0.4, and the RMS residual of 62.90% of the test points is less than 1.0 m.   Conclusions:  This method can obtain the temporal and spatial distribution characteristics of groundwater, and provide data support for the research of groundwater resources.