Application of partial least squares regression in detecting the important landscape indicators determining urban land surface temperature variation

Revealing the interaction between landscape pattern and urban land surface temperature (LST) can provide insight into mitigating thermal environmental risks. However, there is no consensus about the key landscape indicators influencing LST. This study sought to identify the key landscape indicators influencing LST considering a large number of landscape pattern variables and multiple scales. This study applied ordinary least squares regression and partial least squares regression to explore a combination of landscape metrics and identify the key indicators influencing LST. A total of 49 Landsat images of the main city of Shenzhen, China were examined at 13 spatial scales. The landscape composition indicators derived from biophysical proportion, a new metric developed in this study, more effectively determined LST variation than those derived from land cover proportion. Area-related landscape configuration indicators independently characterized LST variation, but did not give much more new information beyond that given by land cover proportion. Shape-related landscape configuration indicators were effective in combination with land cover proportion, but their importance was uncertain when temporal and spatial scales varied. The influence of landscape configuration on LST exists but should not be overestimated. Comparison of numerous variables at multiple spatiotemporal scales can help identify the influence of multiple landscape characteristics on LST variation.