Modelling land surface temperature in urban areas using spatial regression models

Abstract

This study is aimed at modelling land surface temperature (LST) using the spatial lag model (SLM) and spatial error model (SEM). Landsat-8 OLI/TIRS data and digital elevation model were used to generate the dependent and explanatory variables. First, correlation between the variables and LST, and presence of spatial autocorrelation were assessed. Second, the modelled LST was validated. The findings revealed that built-up areas, green areas and water bodies exhibit lower LST compared to the non-urbanized areas around the city. A moderate inverse relationship (r2 = 0.6) is observed between LST and vegetation index with p-value = 4*10−11. In contrast, built-up and surface water indices, albedo, and elevation indicate a weak positive correlation with LST. The LST predicted by SLM ranged between 20 and 42.9 °C (0.5 °C and 0.7 °C below the minimum and maximum of the original data, respectively). In the case of SEM, the predicted LST ranged between 20.4 and 42.2 °C (only 0.1 °C below the minimum of the original data). At 0.01 level of significant, all the variables are significant predictor of the LST except elevation. Both models performed well but SEM showed more superiority. The outcome of this study will enable planners to obtain insight into interventions that are necessary in order to mitigate surface temperature in urban areas.