Improvements in land surface temperature and emissivity retrieval from Landsat-9 thermal infrared data

Land surface temperature (LST) is the key parameter for characterizing the water and energy balance of the Earth’ surface. At present, thermal infrared (TIR) remote sensing provides the most efficient way to obtain accurate LST regionally and globally. Among existing satellites, the Landsat-9 could observe the Earth's surface via two TIR channels, making it possible to generate the global LST product with a remarkable spatial resolution of 100 m. Currently, the single channel method and split window method generally were used to recover LST from the Landsat-9 TIR measurements. However, accurate land surface emissivity (LSE) is needed in both algorithms, which is very difficult to obtain at the pixel scale. To overcome this issue, an improved LST and LSE separation method was proposed in this study. Firstly, the traditional water vapor scaling (WVS) method was refined to address the atmospheric effects in the satellite measurements. Then, the traditional temperature and emissivity separation method (TES) was adapted to the Landsat-9 observations with only two TIR channels. Finally, an iterative process was designed to retrieve the LST and LSE simultaneously. Validations using in-situ measured LST indicated that the root mean square error (RMSE) of the retrieved LST was around 2.92 K, outperforming the official Landsat-9 LST product with an RMSE of about 4.20 K. Taking ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) products as the references, the RMSE of our retrieved LST and LSE was found to be < 1.55 K and