Grassland degradation on the Tibetan Plateau is closely linked to soil water loss, yet the mechanisms connecting bare-soil patch configuration to soil moisture remain poorly resolved. This study integrates UAV-derived patch metrics with high-resolution field measurements of soil water content (1,574 quadrats) to evaluate spatiotemporal controls on water dynamics. Results show that larger bare-soil patches and stronger connectivity amplify radiation- and wind-driven drying, while heterogeneous patch mosaics accelerate lateral loss across ecosystems. Spatiotemporal models, including GLS-AR(1) and random forest, highlight patch configuration as a reliable early-warning indicator of moisture thresholds and incipient alpine grassland degradation. These insights provide a scalable basis for targeted monitoring and adaptive management in fragile alpine ecosystems.