Improvement of mono-window algorithm for land surface temperature retrieval integrated with subpixel mapping for Landsat imagery

Since a large proportion of pixels are often composed of mixed land cover types within remote sensing images, how to eliminate the impact of the error resulting from pixel mixing effect in the estimation of land surface emissivity on the accuracy of land surface temperature (LST) retrieval from remote sensing data is a key problem to resolve firstly in process of LST retrieval. Based on the local relationship between the thermal radiance of one pixel and that of its components satisfying the Planck's radiance function, in this paper, the mono-window algorithm was improved by integrating with mixed-pixel classification and sub-pixel mapping for Landsat imagery. Validation indicates that the improved mono-window algorithm is able to provide more accurate LST than the original algorithm for Landsat TM/ETM+ imagery. By applying the improved algorithm to the Landsat image of Shanghai, the result revealed the spatial heterogeneity characteristics of UHI effect in Shanghai city.