Hourly impact of urban forests on land surface temperature based on machine learning

Urban forests plays a critical role in mitigating the urban heat island (UHI) effect, yet its hourly and structure-dependent cooling performance remains insufficiently understood - particularly in morphology-constrained heritage cities. This study investigates the hourly regulation of land surface temperature (LST) by vegetation structure in Rome, a compact city with limited greening opportunities. We use multi-temporal Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) thermal imagery across 12 clear-sky summer scenes and analyse vegetation–temperature relationships with random forest (RF) models, interpreted through Shapley additive explanations (SHAP) and partial dependence plots (PDPs). The models showed high predictive performance and identified nonlinear thresholds: strongest midday cooling at MTH < ∼6 m, and additional afternoon cooling with canopy cohesion (AIₜ ≈ 70–95). Within UNESCO World Heritage zoning, the results provide structure-informed, hour-specific greening targets and demonstrate an hourly machine-learning attribution of vegetation–temperature relationships.