Evaluation of spatially heterogeneous driving forces of the urban heat environment based on a regression tree model

Abstract

The urban heat environment, represented by land surface temperature (LST), is highly heterogeneous. Previous studies have investigated the universal physical laws during LST modeling. However, spatial patterns of the driving forces and their temporal differences have not been well examined. In this study, spatially heterogeneous effects of two important urban biophysical compositions, Normalized Difference Vegetation Index (NDVI) and impervious surfaces, on LST are determined by a regression tree model. The results showed that LST patterns varied on different dates. Additionally, nonlinear relationships existed between LST, NDVI, and impervious surfaces. The regression tree model outperformed the multiple linear regression in modeling LST variations. Furthermore, the "rules" derived from regression tree indicated the forces that are driving LST locally. These results suggest that the "rules" can be used as urban planning indicators and provide insights into the variations in driving forces linked to the urban landscape. We found that the regression tree is an important approach to adopt for conducting studies on urban heat and provides insight into complex and hierarchical LST patterns and processes, thus facilitating urban heat island mitigation in terms of reducing energy use and air pollution in urban areas.