Estimation of subpixel land surface temperature using Landsat TM imagery: a case examination over a heterogeneous urban area

Land surface temperature is a key surface parameter in urban ecological and environmental studies. This paper took Guangzhou core urban area as the study site and selected Normalized Differential Vegetation Index (NDVI), Normalized Differential Build-up Index (NDBI), Normalized Difference Bareness Index (NDBaI), (Digital Evolution Model) DEM and SLOPE as representative surface characteristics indices in urban, to estimate subpixel land surface temperature (LST). By using correlation analysis and linear regression approaches in multiscale, a subpixel LST estimation model was constructed with the consideration of scale parameter based on Landsat TM. Results found that the multi-scale model preformed a suitable approach in estimating subpixel LST in 90m, 60m and 30m with low RMSE and MAT. Correlation analysis indicated that relationship between LST and those selected surface characteristics indices was scale dependence obviously. Multiple linear regression models of LST and surface characteristics indices changed during scaling. Accuracy of estimation results had highly related to landscape characteristics. For example, Central Business District (CBD) with high density built-up showed higher accuracy compared to the other landscape mainly covered with bare soil. This study may provide a simple but effective method to derive subpixel LST image with higher spatial resolution for studying and monitoring the LST in heterogeneous urban areas.