A three-layer decomposition model is used to downscale the surface temperature of the study area via four different surface feature factors and their combinations. Experimental results show that the three-layer decomposition model using multi-surface feature factor combination obtains higher downscaling accuracy than that using a single factor, and the root mean square error increases from 0.813 K to 0.763 K; the main source of error is the architectural area. To study the urban heat island effect, this study uses the heat field intensity index and heat field variation index for evaluation; both evaluation indexes indicate that the accuracy of the multi-factor model is better than that of the single factor model.