Quantifying the multilevel effects of landscape composition and configuration on land surface temperature

Although prior studies have suggested that both landscape composition and configuration significantly affect land surface temperature (LST), the hierarchical effects of these landscape metrics on LST have been neglected because of the use of single-level models. With the analysis of remote sensing images of Zhuhai city in Southern China, this study incorporates multilevel models to estimate the hierarchical effects of landscape composition and configuration on LST. Comparisons of the single-level ordinary least squares (OLS) regression model and the multilevel models show four findings. First, LST is influenced by both level-one effects of composition (pixels) and level-two effects of configuration (groups of pixels). Second, multilevel models outperform OLS model in accounting for the LST-landscape relationships as the former produces the better fitting results and the smaller residuals and autocorrelations than the latter. Third, configuration metrics account for greater variability in LST than composition metrics. Fourth, the effects of configuration metrics vary across the level-two groups. The findings of this study provide new insights into the landscape influences on LST, and suggest that for mitigating urban heat island effects, optimizing the configurations of land cover types in urban areas should be considered because of the larger cooling effect of landscape configurations than compositions on LST.