Impact mechanisms of 2D and 3D spatial morphologies on urban thermal environment in high-density urban blocks: A case study of Beijing's Core Area

Abstract

Amid global warming, thermal issues in high-density urban areas have become increasingly severe. This study developed an integrated framework combining artificial neural networks, machine learning, and spatial regression to assess how urban morphology impacts the thermal environment in high-density urban areas. Using 2D and 3D building morphology indicators, we applied self-organizing maps (SOM) to classify urban blocks. Boosted regression trees (BRT) model identified key urban morphology indicators influencing land surface temperature (LST), evaluated their contributions, and analyzed marginal effects. Generalized additive models (GAM) explored nonlinear relationships and threshold effects between LST and morphology metrics, while geographically weighted regression (GWR) captured spatial heterogeneity. Results showed that Dense Low-rise Zone (DLZ) and Dense Mid-rise Zone (DMZ) are predominant in Beijing's core area, with DLZ averaging 50.2 °C, significantly impacting the thermal environment. Building patch density (PD_B) and building number density (BND) are primary indicators affecting LST, with 3D indicators showing the strongest influence. ISF, HRE and POID exhibit significant nonlinear relationships with LST. These findings provide a scientific basis for thermal environment planning and management strategies in high-density urban settings, offering guidance for mitigating urban heat island effects.