Alpine arid ecosystems are highly sensitive to climate change, necessitating an understanding of vegetation productivity for ecological management and adaptation. This study investigated net primary productivity (NPP) patterns in the Qilian Mountains, a representative alpine arid region in western China, from 2001 to 2022. Using a localized Carnegie-Ames-Stanford Approach (CASA) model combined with statistical and machine learning methods, we analyzed the spatiotemporal patterns, change trajectories, and climatic response mechanisms of NPP. Results indicate a significant increase (p <0.05) in NPP across 59.90 % of the study area, particularly in the central and western sections. Increased precipitation and glacial meltwater drove grassland expansion and productivity improvement. Temperature emerged as the dominant driver of NPP interannual variations, exerting a stronger influence than precipitation, vapor pressure deficit, and solar radiation across 39.04 % of the region. Trajectory analysis revealed that while positive NPP changes predominantly followed linearity, negative changes were typically characterized by abrupt shifts. The change in moisture availability is regarded as a critical determinant of the instability of NPP. Positive abrupt shifts of NPP often occur in wet years, and low (<15 %) or high (>32 %) precipitation variability as well as reduced solar radiation (slope <-0.9 MJ/m(2)/year) are all conducive to the positive abrupt shifts of NPP. Furthermore, in arid ecosystems (aridity index < 0.22) with fractional vegetation cover below 14 %, the likelihood of positive abrupt shifts is notably higher. These findings provide critical insights for ecosystem conservation strategies and predicting ecological responses to future climate scenarios in vulnerable alpine arid regions.