Revealing the impact of urban land use patterns on land surface temperature through graph attention networks

A considerable body of literature focuses on the intensification of urban heat island effects (UHI) due to urbanization. However, the impacts of interactions between parcels on land surface temperature (LST) remains unclear. In this study, Graph Attention Networks model represented the land use (LU) pattern as a graph structure to quantify the interaction between parcels, and further investigated its impact on LST. The results elucidated the strong influence of LU on LST not only comes from itself but also adjacent parcels. Roadway land, industrial/storage land, and water significantly affect the LST of adjacent parcels. In contrast, administrative land, park/green space, arable land, forest land, and grassland had a stronger impact on themselves. This strong influence varies with the topological distance (namely, order). For instance, The influence of commercial land on the focal parcel' LST within the four orders increases with the rise in topological distances. Finally, it was further clarified which types of parcels were more affected by specific LUs. These findings would provide the implementation of the LU layout to mitigate the UHI and promote sustainable urban development.