An Efficient Approach for Pixel Decomposition of Land Surface Temperature from Landsat TM Data

The spatial resolution of land surface temperature (LST, 120 m) retrieved from thermal infrared (TIR) band is lower than its visible/near-infrared (VNIR) bands (30 m). LST image with high spatial resolution compatible with VNIR bands of Landsat TM is very important for application of the LST image to many studies such as environmental monitoring. The objective of this study is to decompose the coarse pixels of the LST image data into the same pixel scale of the auxiliary VNIR data. Firstly, the E-DisTrad method is used to divide LST of the parent pixels into the sub-pixels to obtain the first decomposition temperature and the theoretical radiance at-sensor of each sub-pixel. Then, a chess-segmentation is done to the temperature of the sub-pixels on the basis of the object-oriented image segmentation method to compute the weight for each sub-pixel, which is consequently used to allocate the thermal radiance for each sub-pixel to generate the decomposed LST image. At last, a method of increase the spatial resolution firstly and then used for double-step pixel decomposition is executed in the study to validate the accuracy and efficiency of the decomposition method for the Landsat TM image of Beijing. The result showed that the double-step pixel decomposition method is applicable for decomposition of LST images for high spatial resolution, reflecting the spatial differences of different land use and land cover and ensure the conservation of energy before and after pixel decomposition. Therefore we can conclude that it is ideally suited for complex covered surface area of downscaling thermal infrared band remote sensing data.