Exploring the effects of street canyon morphology on LST within different street types using causal inference and machine learning

Abstract

There is currently a lack of classification methods for street canyon morphology at the street-scale level. This can impede the development of targeted cooling strategies tailored to the specific characteristics of different street morphologies. This study quantifies street canyon morphology using street-view hemisphere images and compares multiple clustering models to identify the optimal model and parameters. Subsequently, machine learning is coupled with causal inference models to explore the associative mechanisms between different street canyon morphology indices and multi-time land surface temperature (LST). The results reveal that spectral clustering divides streets into three categories of wide streets and two categories of narrower alleys. Different street types exhibit distinct correlation trends with LST, highlighting the importance of clustering algorithms. In conjunction with the results of causal inference, it is observed that alleys with high canopy coverage and broad streets equipped with road-center hedges demonstrate superior cooling capabilities, with cooling effects of 23.56 % and 18.81 %, respectively. Conversely, for broad streets with lower levels of greening, increasing the height of roadside buildings can be an effective strategy to maximize the utilization of building shadows and wind for cooling purposes. This study emphasizes vegetation as a key factor in altering street canyon morphology to achieve cooling effects, particularly in stock developments.