Unveiling the 3D Resilience A Deep Learning Approach to Quantify Urban Green Space Patterns in Buffering Extreme Heat

Urban green infrastructure plays a critical role in mitigating thermal stress and enhancing climate resilience, yet traditional two-dimensional metrics (NDVI) fail to capture the vertical structural complexity that determines cooling performance. This study introduces a computer vision approach using monocular depth estimation to extract three-dimensional green infrastructure metrics from street view imagery, democratizing 3D urban analysis previously constrained by expensive LiDAR systems. By integrating depth information with semantic segmentation, we calculate Estimated Green Volume Density (EGVD) and assess its relationship with thermal resilience across 487 urban locations. Our Random Forest analysis, controlling for Building View Factor (BVF) as a proxy for building density, reveals that EGVD proved to be a stronger predictor (importance:[[EQUATION]]) than Building Density (importance:[[EQUATION]]), demonstrating that three-dimensional vegetation structure contributes independently to thermal resilience beyond neighborhood characteristics. We identify five distinct green space patterns, with Pattern 1 (High-Canopy-Density Tree Corridor) exhibiting superior thermal resilience ([[EQUATION]]). A critical cooling threshold at EGVD ≈ 0.75 m3/m2 (75th percentile) reveals non-linear relationships where locations above this threshold show enhanced resilience gains. Our stratified analysis demonstrates that optimizing vertical structure yields significant additional resilience gains independent of socioeconomic context, providing actionable insights for evidence-based urban green infrastructure planning.These findings confirm that vertical structure is an actionable intervention point independent of neighborhood morphology.