关键词: 城市扩张, 树冠分割, 深度学习, 卫星影像, 生态恢复

Abstract: Large-scale terrain reshaping associated with rapid urban expansion has become an increasingly important driver of ecological change in fragile environments, yet its impacts on forest structure remain poorly quantified at the tree level. Using the world's largest mountain excavation urbanization project, the Mountain Excavation and City Construction (MECC) project in Yan’an, China as a representative case, this study investigates how intensive anthropogenic terrain modification reshapes forest spatial structure before, during, and after urban construction. To enable tree-level assessment of forest dynamics under highly heterogeneous land-cover conditions, we developed an improved deep-learning approach, adaptive tree crown segmentation (ATCS), to extract individual tree crowns using 0.5 m satellite imagery. Tree-level results reveal that crown number and total crown area declined by 83.9 % and 81.5 % during the main construction phase and recovered to 68.4 % and 56.3 % of preproject levels after construction. However, crown density and cover remained low in urban core areas, with trees increasingly distributed in fragmented and low-density patches. Regression analysis further revealed strong negative relationships between forest structural metrics (crown density and cover) and proxies of urban intensity (nighttime light intensity and population density). These findings indicate that large-scale terrain reshaping causes persistent alteration in forest spatial structure that is not fully mitigated by post-construction greening.

Key words: Urban expansion, Tree crown segmentation, Deep learning, Satellite imagery, Ecological restoration