Sustainable management strategies for water resources rely on accurate knowledge about the dynamics of hydrological processes, especially in drylands, where freshwater is the limiting factor for the development of human society and ecosystems. The populated Loess Plateau (LP) in North China is a typical semi-arid region where competition for water between people and nature is worth noting because of afforestation promoted by the Grain to Green Program. In this study, changes in key components of terrestrial water storage (TWS) in the LP were explored using a multi-satellite approach, including Gravity Recovery and Climate Experiment (GRACE) observations and Earth observations of precipitation, evapotranspiration and soil moisture. By integrating data on human water use from different sources with satellite observations, we were able to examine the mechanisms driving these changes. The results demonstrated that, according to an evaluation based on reproducing TWS computed from the regional water balance in the LP, the mascon solution of the Center for Space Research (CSR) at University of Texas at Austin performed best out of the commonly used GRACE products. Regional TWS derived from the CSR mascon solution in the LP decreased significantly for the period 2003-2015. Significant decreases were also detected for regional ground water storage (GWS) estimated by decomposing the GRACE TWS using multi-sources remote sensing data. GWS made the greatest contribution to the changes in TWS. Increased plant transpiration was one reason for the decreasing trend of GWS. Because changes in precipitation, soil moisture and water consumed by irrigation were minor at regional scales, we concluded that the increase of transpiration is driven by deep-rooted trees planted, which use the part of precipitation that previously recharged groundwater. The findings from this study are valuable for water resource management and ecological restoration in semi-arid regions with high populations.