Spatiotemporal Patterns and Associated Factors of Rural Shrinkage in the Jinsha River Basin, Yunnan Province: An Interpretable Ensemble Learning Approach

Rural communities shrink when residents leave, farmland and housing are underused, and local economic and service functions weaken. This study examines where this process occurred in 545 townships in the Yunnan section of the Jinsha River Basin and which local conditions were most closely associated with it in 2000, 2010, and 2020. We combined population, land, and economic indicators with spatial analysis and an ensemble machine-learning model designed to explain how each factor contributed to prediction. Rural shrinkage was widespread in 2000, eased substantially by 2010, and became spatially polarized again by 2020. Spatial clustering weakened and then strengthened across the three years. Coordination among population, land, and economic subsystems improved overall but remained uneven. Public-service accessibility was most prominent in the 2000 model, whereas vegetation conditions and distance to the provincial capital were more prominent in 2020. The results identify priority locations for monitoring, clarify how environmental, accessibility, and service conditions combine in different places, and provide a transferable framework for diagnosing rural shrinkage while separating robust spatial evidence from policy options that still require local testing.