Groundwater plays an important role in ecological systems. However, it is difficult to investigate the spatiotemporal changes in groundwater storage (GWS) in the arid areas of Northwest China (NC) due to the limited observation data. Selecting the Tarim River Basin (TRB) in NC, this paper analyzed the variations in GWS and its relationships with climate factors based on downscaling the GRACE data (1 km × 1 km). We used random forest regression (RFR) to establish a model for GWS downscaling simulation by the grid data of GRACE-derived GWS, vegetation index and temperature. We also tested the validity of the downscaled results by comparing the errors between the downscaled GWS and the observed groundwater level by statistical indexes. The results showed that the GWS in the TRB mainly presented decreases from 2003 to 2016, with a significant decline in the southern Tianshan Mountains. The decreasing amplitude of GWS gradually weakened from north to south, and the fluctuations in the mountains were stronger than those in the central basin. The GWS changes presented seasonal differences, with strong decreases in summer and autumn but slight decreases in spring and winter. We also found that GWS in the TRB was more affected by temperature. The impact of precipitation on GWS changes was relatively weak, while temperature and evaporation had relatively significant influences on GWS changes. The negative effect of temperature on GWS in mountainous areas was still noticeable after a lag time of 1–4 months. The development of agricultural land and groundwater exploitation affected GWS changes.