Greening the concrete jungle: Unveiling the co-mitigation of greenspace configuration on PM2.5 and land surface temperature with explanatory machine learning

Abstract

In densely populated urban areas, the amount of urban green space (UGS) is limited and increasing it can be challenging due to the high proportion of unvegetated land. Therefore, it is crucial to determine the optimal spatial configuration of UGS to achieve environmental benefits such as reducing PM2.5 concentrations and land surface temperature (LST). However, there is limited research on this topic. This study employed an explanatory machine learning method to identify the non-linear relationships between contributing factors and the co-mitigation of PM2.5 and LST in highly-dense urban areas, which had three key advantages: improved accuracy, spatially explicit information, and enhanced understanding of complex relationships. The study found that maintaining a UGS proportion of 25–30 % was desirable for mitigating PM2.5 and LST. Additionally, maintaining an aggregation index above 97, a patch density above 1650, and a largest UGS patch proportion between 2.00 % and 4.85 % was beneficial for co-mitigation. However, the study found conflicting results between shape complexity and co-mitigation. Surprisingly, higher road density appeared to mitigate both PM2.5 and LSTs. Overall, the study highlights the potential of explanatory machine learning methods for sustainable urban environmental management, providing insights into the co-mitigation effects of UGS and urban morphology.