Explainable Machine Learning Reveals Parameter Thresholds and Scene-based Adaptation of Tree Configurations

• Built a teacher–student pipeline that fuses ENVI-met simulation outputs with gradient-boosted models to predict thermal comfort from 3D tree structures at street scale.
• Integrated LiDAR and inventory data to derive tree configurations and minimal urban-form controls for robust generalisation across districts, climates, and street geometries.
• Produced variable importances, practical thresholds, and interaction effects to generate context-specific design guidance for optimising tree configurations.