A new Emissivity Retrieval Method for Landsat

Landsat data are the important sources for the inversion of land surface temperature (LST) with high spatial resolution. As an important parameter of LST inversion, the land surface emissivity (LSE) of Landsat TIR channels is usually inverted by the semi-empirical method, which has certain limitations. With the launch of Landsat 9, the requirement for LSE became more urgent. Therefore, this paper proposed a new retrieval method to estimate LSE with high spatial resolution for Landsat. Reflectance of Landsat VNIR-SWIR channels were used for the LSE inversion by Gradient Boost Regression Tree (GBRT) machine learning method. In order to evaluate the accuracy of the estimated results, the LSEs were compared with those estimated by NDVI threshold method and the field measurement data. The results showed the LSEs estimated by GBRT model were consistent with the measured data. So it demonstrated that this method was feasible to estimated LSE for Landsat.