Monitoring and understanding of groundwater variability is critical for water resource management but remains highly challenging in vulnerable arid regions with limited ground observations. This study explores improved characterization for groundwater storage anomalies (GWSA) and recharge dynamics by leveraging GRACE observations, land surface models, key climate variables, and in-situ well records, combined with complementary modeling strategies. Multiple GRACE-derived terrestrial water storage anomaly (TWSA) solutions were examined to support a reliable observational basis for the analysis. The GRACE-Noah model indicated that the long-term and seasonal trends in GWSA are highly synchronized with TWSA, emphasizing groundwater's central role in terrestrial water storage, while snowmelt was also found to influence groundwater dynamics through its impact on surface water and soil moisture. The well-configured artificial neural network (ANN) model demonstrated superior performance, proving to be effective for capturing connections between key hydroclimate factors and groundwater. Results revealed a significant decline averaging -4.34 mm/yr in groundwater storage since 2002 (-4.51 mm/yr from GRACE-Noah and -4.17 mm/yr from GRACE-ANN), highlighting ongoing stress on groundwater resources. The spatiotemporal variations in groundwater recharge and depletion across basins are characterized, revealing the depletion patterns around the Tianshan Mountains and seasonal dynamics driven by snowmelt and evapotranspiration. By integrating GRACE observations with complementary modeling strategies, this study contributes to improving the characterization of groundwater depletion and recharge dynamics in Northwest China, advancing knowledge of groundwater system behavior under changing climatic conditions, further providing new insights for effective groundwater resource assessment in arid regions.