Slow-moving landslides are widespread in alpine canyon regions and pose a major threat to transportation corridors. Affected by active tectonics, steep terrain, and hydroclimate, the relationships between environmental conditions and landslide occurrence are strongly nonlinear and spatially heterogeneous. Such inherent spatial variations greatly limit the applicability of conventional landslide analysis methods based on the spatial stationarity assumption, making it difficult to accurately capture the spatial differentiation laws of landslide development. In this study, firstly, along the Linzhi–Ya’an section of the Qinghai–Tibet Plateau Transportation Corridor (LY-QTPTC), the GACOS-assisted Interferometric Synthetic Aperture Radar (InSAR) was applied to derive annual displacement rates. Then, slow-moving landslides were detected through a deformation gradient–based extraction constrained by Slope Units (SUs), and field surveys were conducted to validate the detected landslide inventory. Finally, an interpretable machine learning model combined with GeoShapley was further developed by integrating geographic effect (GEO) and conditioning factors to quantify spatial effects and variable interactions. A total of 3,861 slow-moving landslides were detected, exhibiting pronounced spatial clustering patterns. GeoShapley interpretation revealed that temperature, precipitation, GEO, and NDVI were the four most important conditioning factors, contributing 13.27%, 10.54%, 8.60%, and 7.00%, respectively. The effects of conditioning factors are not spatially stationary across the study area, instead, temperature, precipitation, and NDVI exhibit strong spatially differentiated driving effects, with markedly different patterns across major landslide concentration zones. In particular, identical environmental conditions may correspond to contrasting landslide responses under different geographic settings, highlighting the importance of spatial context in governing slope failure. The proposed approach enhances the reliability of regional landslide detection and provides new insights into spatially varying landslide controls in tectonically active alpine canyon regions.