In the context of climate change, the hydrological processes and water resource system vulnerabilities in inland river basins of arid regions are intensifying. Understanding their evolutionary patterns and driving mechanisms is crucial for sustainable water resource management, agricultural development, and the protection of ecological security. This study focuses on the Kaidu River Basin, systematically analyzing the temporal and spatial variations in hydrological cycle elements in the basin from 1998 to 2023 based on multi-source precipitation data, the SWAT hydrological model, and the glacier degree-day model. The study also identifies the main driving factors using a geographic detector. The results show that the SWAT model performs well (calibration period R2 and NSE >= 0.75, validation period R2 and NSE of 0.75 and 0.70, respectively), indicating reliable simulation results. The surface water resources and the contribution of glacier meltwater to runoff in the basin both show a fluctuating downward trend, while potential evapotranspiration increases. The contribution of glacier meltwater during the ablation season decreased from 69.86% in 2014-2016 to 45.01% in 2017-2021. The hydrological processes exhibit a spatial pattern of "mountain areas generating runoff, non-mountain areas consuming water". The geographic detector results indicate that precipitation is the decisive factor for the spatial differentiation of hydrological processes (influence degree q = 56.9%), with temperature, potential evapotranspiration, and altitude playing important synergistic roles. Moreover, the explanatory power of multi-factor interactions is much greater than that of individual factors. The findings of this study provide a scientific basis for the optimized allocation of watershed water resources, efficient agricultural irrigation, and the sustainable development of oasis ecosystems under changing environmental conditions, thereby supporting the goals of water security and sustainable development in inland river basins of arid regions.