Urban heat island is a commonly observed phenomenon and an actively researched topic. One of the contributors to urban heat islands is the absorption and trapping of heat from complex three-dimensional building structures in megacities. Studies have been looking at the relationship between satellite observed land surface temperature and the type of land use/urban category data. However, the complex three-dimensional building information is always simplified in those datasets, whereas in-situ measurement is always limited in spatial coverage, hindering the study on regional impact. In this project, open-source machine learning algorithms will be used to study the relationship between the urban surface temperature observed from satellite images and the three-dimensional urban structure in the greater bay area. The machine learning model would be trained to estimate the land surface temperature and compared to the one estimated by regression analysis.