Beyond Green vs. Gray: A Multi-Scale Machine Learning Analysis of Urban Morphology and Topography on Land Surface Temperature

Urban heat island (UHI) mitigation strategies frequently rely on two-dimensional land-cover metrics, often failing to account for the complex, non-linear interactions between built morphology and natural terrain. This study investigates the scale-dependent thermodynamic drivers of land surface temperature to establish empirically verified thresholds for urban zoning. Using the topographically heterogeneous city of Cincinnati, Ohio, as a case study, a Random Forest machine learning ensemble, Spearman rank-order correlations, and segmented regression were applied to high-resolution geospatial and architectural datasets. Thermal dynamics were analyzed across discrete micro- and macro-scale grids to identify specific biophysical tipping points. The RF model achieved high predictive accuracy. Testing R² values improved from 0.817 at the micro-scale to 0.833 at the macro-scale. The results highlight a major scale-dependent inversion in UHI drivers. At the pedestrian micro-scale, three-dimensional Building Volume Density acts as the primary thermal predictor (59% variable importance). At the neighborhood macro-scale, thermodynamic behavior shifts back to a two-dimensional paradigm. Here, aggregate vegetation fraction (31.5%) and building footprint density (25.5%) dominate. Segmented regression extracted precise biophysical thresholds. The data shows that morphological features can invert their thermal functions based on the scale of aggregation. Average street aspect ratio provides localized pedestrian shading above a micro-scale breakpoint of 2.46. However, it acts as a regional longwave radiation trap when aggregated beyond a macro-scale threshold of 1.76. Building density also exhibits a thermal saturation plateau. This occurs at 40.2% micro-coverage and 8.57% macro-coverage, where mutual self-shading caps additional heat accumulation. Model simulations disproved a hypothesized topographic penalty (p = 0.38). Instead, the data reveals that low-volume, sprawling urban typologies cause vegetative canopies to trap outgoing surface radiation. Urban thermodynamic drivers are inherently scale dependent. Traditional footprint-restricted zoning is insufficient for effective heat mitigation. To engineer climate-resilient cities, municipal planning frameworks must transition to multi-scale, topography-aware regulations that manage 3D volumetric mass using established geometric thresholds.