高密度城市中的形态敏感洪水韧性：基于图的分区与可解释机器学习方法

Urban flood resilience assessment in high-density cities traditionally adopts generalized indicator frameworks based on the spatial homogeneity assumption, which weakens the capture of localized resilience mechanisms and the formulation of targeted adaptation strategies. This study develops a morphology-sensitive flood resilience framework that incorporates urban spatial heterogeneity. Taking ultra-high-density coastal Macau as the case, it integrates graph-based morphological zoning and explainable machine learning to examine spatial variations in resilience mechanisms. Results show flood resilience is morphology-dependent rather than spatially uniform. Four morphological zones present distinct resilience patterns, and conventional frameworks have inconsistent applicability and cause evident assessment bias. Nonlinear modeling identifies threshold effects and inter-factor interactions of built-environment variables, with resilience responding nonlinearly to infrastructure, vegetation and urban form. This research advocates a shift from aggregate unified evaluation to zone-specific resilience planning, providing a transferable approach and practical references for climate adaptation and differentiated flood mitigation in high-density cities.